Data center new builds diminish even as demand rises

New findings from commercial real estate services and investment firm CBRE Group reveal a dichotomy in the vibrant data center market across North America.

At the same time as the sector set records in 2025 for overall activity, new facilities were not being built, not because demand was slowing, but due to the difficulty in obtaining construction permits and power. CBRE also observed that community involvement is becoming a key driver in approvals.

“The size and scale of new development projects are now reaching 100+ acres for a greenfield campus,” noted Pat Lynch, executive managing director, CBRE Data Center Solutions. “Ongoing partnership and good working relationships with local municipalities who seek the benefits of increased assessed value for annual property tax, thousands of construction jobs, reoccurring sales tax on equipment refreshes, low impact on schools/vehicular traffic are important for data center developers and operators.”

Yet, despite this record demand, construction activity declined for the first time since 2020 due, CBRE explained, “to extended timelines tied to permitting, zoning approvals, and sourcing adequate power.” Because of this, it said, many projects remain stalled at the planning stage; capacity under construction fell to 5,994.4MW at year end, a decline of 5.5% compared to 2024.

CBRE’s North American Data Center Trend Report also found that:

• Primary market supply increased by 36% year-over-year to 9,432MW, which surpassed 2024’s 34% increase due to accelerated hyperscale demand.
• The overall data center (DC) vacancy rate in primary markets fell to a record low of 1.4% at the end of last year, despite the increase in capacity.
• Average monthly rental rates for requirements between 250kW and 500kW rose by 6.5%, the fourth consecutive annual increase, as supply constraints generated upward pressure on prices.
  For 3MW-10MW, the price surged by 12.5%. This trend is expected to continue, exceeding the rate of inflation for the next two to five years.
• Facilities optimized for AI (with, for example, liquid cooling and high power density racks) commanded rental premiums over conventional colocation facilities.
• In most markets, grid power capacity is “largely booked” through 2030, so new projects are looking at onsite power generation such as natural gas generators, wind turbines, hydrogen fuel cells, or solar panels paired with batteries. Small modular nuclear reactors, the report said, are expected to become practical as early as 2035.
• Inference AI is “redefining data center demand with a need for more regional and distributed data centers,” generating a requirement for low latency, scalable infrastructure.

“New markets for development continue to defy the historical norm of data center locations,” Lynch observed. “Historically, data centers typically had to be clustered within a 25-mile radius around an internet exchange point (IXP) or carrier hotel. Due to improvements in fiber connectivity and capacity, we are seeing development steer towards more rural areas where there are hundreds of acres of available space.”

However, the report said, development in more remote regions “will remain challenging” due to a shortage of skilled labor such as mechanics, electricians, plumbers, laborers and construction workers.

Market shift from abundance to constrained

Sanchit Vir Gogia, chief analyst at Greyhound Research, said Wednesday that enterprises must assume, as the report suggests, that there will be elevated pricing for North American data center capacity through at least 2029, and possibly longer. “Vacancy at or near 1%- 2% is not a temporary imbalance,” he said. It is a “signal that supply elasticity has broken. When over 90% of capacity under construction is already pre-committed, new entrants are negotiating from a position of structural scarcity, not market equilibrium.”

“Energy intensity is rising because AI workloads are more power dense,” he pointed out. “So even if an enterprise does not expand its footprint, the cost per deployed workload can still increase because the electrical envelope changes.”

His advice to enterprises: expansion is viable, but only if they diversify beyond legacy Tier 1 hubs, secure long term expansion rights early, negotiate structured pricing protection, and “optimize workload placement with ruthless clarity.”

But, he added, “it is not viable if enterprises assume that incremental megawatts will remain readily available in the same region at roughly similar economics.”

John Annand, practice lead at Info-Tech Research, said that, to compensate, his firm’s client base is increasingly open to moving the right workloads to private clouds or on-premises. “The shift is nuanced, not ideological,” he said, and is usually financially motivated and “framed as hybrid optimization, not public cloud reversal.”  

With the constraints on power availability, and on regional willingness to authorize building permits, he said, “there is huge competition for increasingly scarce opportunities to build data center capacity.” 

He pointed out that hyperscalers and colocation/wholesaler firms (Amazon, Microsoft, Equinix, and others) make up the bulk of those doing data center construction, and much of the colocation capacity is already under contract to be leased back to the hyperscalers.

So, he said, due to fewer builders, low vacancy rates, and considerable pre-leasing of planned capacity, enterprises that don’t have the leverage of the big five should expect their rates to rise. “[Contract] terms with 5% annual escalators and cost pass-through when the [electrical] grid needs to be upgraded are commonplace,” he noted. 

“It’s always a question of economics,” he added. “The debt taken out by these builders will have to be serviced by their customers, and the operators will, of course, want to make a profit. If you’re in a position to strategically invest your company’s capital to build out your own capacity, especially if the nature of your company means it has some of the necessary precursors like facilities, redundant power, and telecommunications, maybe it is a good time to pretend it’s the 2010s again!” 

The bottleneck is power, AI is the culprit

“The surge in leasing across North America reflects how quickly business and consumers are adopting AI-powered tools and digital services,” Lynch noted. “Demand is outpacing supply, while power and supply chain shortages are reinforcing a power-first approach that prioritizes sites with the fastest path to power. Ultimately, unlocking additional supply will depend on power availability timelines, approvals for on-site generation, and greater investment in transmission infrastructure.”

Gogia agreed. “Construction slowed because physical infrastructure realities have overtaken financial capital as the dominant gating factor,” he said. With the dramatic lengthening of power interconnection queues over the past decade, what once took less than two years from grid connection request to commercial operation has now expanded to four to five years or more. That timeline expansion alone “disrupts traditional development cycles.”

Overall, he said, the market has “shifted from abundant to constrained. Expansion is possible; casual expansion is not.”

This article originally appeared on NetworkWorld.